On 14/09/2022 13.52, Richard Henderson wrote:
From: Peter Maydell <peter.mayd...@linaro.org>
The PMU cycle and event counter infrastructure design requires that
operations on the PMU register fields are wrapped in pmu_op_start()
and pmu_op_finish() calls (or their more specific pmmcntr and
pmevcntr equivalents). This includes any changes to registers which
affect whether the counter should be enabled or disabled, but we
forgot to do this.
The effect of this bug is that in sequences like:
* disable the cycle counter (PMCCNTR) using the PMCNTEN register
* write a value such as 0xfffff000 to the PMCCNTR
* restart the counter by writing to PMCNTEN
the value written to the cycle counter is corrupted, and it starts
counting from the wrong place. (Essentially, we fail to record that
the QEMU_CLOCK_VIRTUAL timestamp when the counter should be considered
to have started counting is the point when PMCNTEN is written to enable
the counter.)
Add the necessary bracketing calls, so that updates to the various
registers which affect whether the PMU is counting are handled
correctly.
Signed-off-by: Peter Maydell <peter.mayd...@linaro.org>
Reviewed-by: Richard Henderson <richard.hender...@linaro.org>
Message-Id: <20220822132358.3524971-4-peter.mayd...@linaro.org>
Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
---
target/arm/helper.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 45 insertions(+)
Hi Peter, hi Richard,
this seems to break some Avocado based test(s) in our CI:
make check-venv
./tests/venv/bin/avocado run
tests/avocado/replay_kernel.py:ReplayKernelNormal.test_aarch64_virt
... fails with commit 01765386a88868ae993bcb but still passes
with the preceeding commit. Could you please have a look?
Thanks,
Thomas
